On Mon, Nov 25, 2019 at 06:12:56PM +0100, Felix Fietkau wrote:
> On 2019-11-25 17:59, Stanislaw Gruszka wrote:
> > On Mon, Nov 25, 2019 at 03:07:59PM +0100, Markus Theil wrote:
> >> On 11/25/19 2:00 PM, Stanislaw Gruszka wrote:
> >> > On Thu, Nov 21, 2019 at 06:59:57PM +0100, Markus Theil wrote:
> >> >> +void mt76x02_mac_set_beacon_finish(struct mt76x02_dev *dev)
> >> >> +{
> >> >> + mt76_wr(dev, MT_BCN_BYPASS_MASK,
> >> >> + 0xff00 | ~bitrev8(dev->beacon_data_mask));
> >> > Since you arrange beacon slots continues starting from 0
> >> > (i.e. 0,1,2 instead of "random" vif_idx values like 0,4,6),
> >> > I think it would make sense to keep
> >> > MT_MAC_BSSID_DW1_MBEACON_N = bcn_idx - 1 and set mask unchanged.
> >> >
> >> > But no strong opinion here, code with bitrev8 looks fine too.
> >> I'd like to keep the bitrev8 code, as it saves a copy over usb for usb
> >> devices, if MT_MAC_BSSID_DW_BEACON_N is kept constant.
> >> bitrev8 should be a rather cheap operation compared to a copy over some
> >> form of bus.
> >
> > This make sense. I tested the code on MT7630E and after adding missed
> > write_txwi function, it works fine. So I think bitrev8 code is ok.
> I find the use of bitrev8/ffz a bit convoluted. If I understand the code
> right, wouldn't it be equivalent to keeping beacon_data_count instead of
> beacon_data_mask and doing:
> mt76_wr(dev, MT_BCN_BYPASS_MASK,
> 0xffff & ~((1 << dev->beacon_data_count) - 1));
> If so, I would strongly prefer the beacon_data_count variant. It's also
> cheaper to calculate, though that probably doesn't matter, since it's
> not really a hot path.
Yes, with new code simple count can be used instead of mask. I did not
pointed this, as it can be further optimization on top of Markus
patches.
Stanislaw
